multi-scale thermal modeling of glass interposer for mobile electronics application › sites ›...
TRANSCRIPT
Multi-scale thermal modeling ofglass interposer for mobileelectronics application
Sangbeom ChoWoodruff School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, Georgia, USAVenky Sundaram and Rao Tummala
3-D Systems Packaging Research Center, Georgia Institute of Technology,Atlanta, Georgia, USA, and
Yogendra JoshiWoodruff School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, Georgia, USA
AbstractPurpose – The functionality of personal mobile electronics continues to increase, in turn driving thedemand for higher logic-to-memory bandwidth. However, the number of inputs/outputs supported bythe current packaging technology is limited by the smallest achievable electrical line spacing, and theassociated noise performance. Also, a growing trend in mobile systems is for the memory chips to bestacked to address the growing demand for memory bandwidth, which in turn gives rise to heat removalchallenges. The glass interposer substrate is a promising packaging technology to address theseemerging demands, because of its many advantages over the traditional organic substrate technology.However, glass has a fundamental limitation, namely low thermal conductivity (~1W/mK). The purposeof this paper is to quantify the thermal performance of glass interposer-based electronic packages bysolving a multi-scale heat transfer problem for an interposer structure. Also, this paper studies thepossible improvement in thermal performance by integrating a fluidic heat spreader or vapor chamberwithin the interposer.Design/methodology/approach – This paper illustrates the multi-scale modeling approach appliedfor different components of the interposer, including Through Package Vias (TPVs) and copper traces.For geometrically intricate and repeating structures, such as interconnects and TPVs, the unit celleffective thermal conductivity approach was used. For non-repeating patterns, such as copper traces inredistribution layer, CAD drawing-based thermal resistance network analysis was used. At the end,the thermal performance of vapor chamber integrated within a glass interposer was estimated byusing an enhanced effective thermal conductivity, calculated from the published thermal resistancedata, in conjunction with the analytical expression for thermal resistance for a given geometry of thevapor chamber.Findings – The limitations arising from the low thermal conductivity of glass can be addressed byusing copper structures and vapor chamber technology.Originality/value – A few reports can be found on thermal performance of glass interposers.However thermal characteristics of glass interposer with advanced cooling technology have notbeen reported.Keywords Multi-scale, Thermal modelling, Interposer, Vapour chamberPaper type Research paper
International Journal of NumericalMethods for Heat & Fluid Flow
Vol. 26 No. 3/4, 2016pp. 1157-1171
©Emerald Group Publishing Limited0961-5539
DOI 10.1108/HFF-09-2015-0378
Received 16 September 2015Revised 13 January 2016
Accepted 13 January 2016
The current issue and full text archive of this journal is available on Emerald Insight at:www.emeraldinsight.com/0961-5539.htm
This research was supported by the Glass and Silicon Interposer Industry Consortium at theGeorgia Tech 3-D Systems Packaging Research Center.
1157
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
Nomenclaturea width of vapor chamberA areaAcopper total copper pixel areaApolymer total polymer pixel areaArow/column pixel area for row/column
thermal resistanceAs heat source areaAtotal total pixel areaA0, Am, An, Amn Fourier coefficientsb length of vapor chamberc width of heat sourced length of heat sourceh heat transfer coefficientk thermal conductivitykcopper copper thermal
conductivitykeff effective thermal
conductivitykin-plane, kx-y in-plane (x-y direction)
thermal conductivitykout-of-plane, kz out-of-plane (z direction)
thermal conductivityLtotal total pixel length
m, n indices for summationQ heat flow rateR thermal resistanceR1D one-dimensional thermal
resistanceRs spreading resistanceRtotal total thermal resistanceRvc vapor chamber total
thermal resistanceAs heat source areat1 vapor chamber thicknessTf condenser cooling
temperatureXc,Yc heat source centroidβm,n, δn, λm eigen valuesyjb junction to board thermal
resistancey mean temperature excess
(T�Tf)ϕ spreading functionς dummy variable
1. IntroductionMobile electronics are packing more features than ever before, and require higherbandwidth (10-30 GB/s) and larger memory capacity, with the same or lower powerconsumption. Additionally, the form factors of such devices continue to shrink, especiallyin thickness. In order to address these demands at a reasonable manufacturing cost,breakthroughs in packaging technologies are needed. 3D integrated circuit (IC) structureswhere multiple chips or dies are stacked have been considered to be an efficient solutionto achieve these goals, but there still remains a number of technical challenges because ofthe large power density. Hence, interposers are considered as a good alternative forstacked integration technology. Interposer is a packaging platform with high density ofelectrical connections, and it is used to fan out the electrical connections to a wider pitch,and also to route the signals between different components placed on the sameinterposer. Because of the high wiring density, interposers can support large number ofinput/outputs (I/Os) required by the advanced IC technology nodes. Figure 1 shows theschematic of a so-called 2.5D glass interposer structure, where two dies are connected toeach other through copper traces in the redistribution layer (RDL) on the interposersubstrate, while both dies are connected to printed circuit board (PCB) through thecopper plated through package vias (TPVs).
Silicon and glass are two major candidates for an interposer substrate. Siliconinterposer has been developed to overcome the limitations of organic substrates due tomany advantages, including high I/O density. However, it is limited to 300 mm wafersizes, leading to high fabrication cost per interposer. Also, its high electrical losses dueto its higher conductivity, limit its electrical performance. To address these issues glass
1158
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
interposers are being developed (Sukumaran et al., 2010). Glass has the advantage ofpanel-based processing, which results in lower cost per interposer. Combined with theadvantages of ultra-high electrical resistivity, and low electrical losses, it becomes anexcellent interposer candidate, especially for mobile applications (Sundaram et al.,2014). However, glass has a thermal conductivity about 100 times lower than that ofsilicon. This limitation can potentially be overcome by incorporating copper structuresand thin heat spreaders for the enhancement of thermal performance. A number ofstudies have investigated the thermal characteristics of interposers through simulation(Lau and Yue, 2009; Zhang and Zhang, 2015). Also, a few experimental thermalcharacterizations of interposer structures have been reported, and compared withsimulations (Oprins et al., 2011; Zhang et al., 2014). Heinig et al. (2014) presented thermalanalysis and optimization results for various 2.5D and 3D integrated processorconfigurations. These results indicated that maximum total power of the processor on25 mm× 16 mm interposer can be increased up to 10W when there is convective heatremoval on bottom side with an effective heat transfer coefficient of 50W/m2 K.Nonetheless, most of the previous works focus on silicon-based integrationtechnologies and thermal studies on glass-based integration technologies arecurrently lacking (Cho et al., 2013; Oprins and Beyne, 2014).
This paper is organized into three parts. The first part explains multi-scale modelingchallenges in interposer modeling and explains the way TPVs, bumps and microbumpsare modeled. The second part describes ways to simplify the copper traces in RDL andPCB for computational efficiency, and compares the effect of different copperstructures on the thermal performance of glass and silicon interposers. In the third part,a vapor chamber integrated glass interposer is introduced and its thermal performanceis characterized. The results from the vapor chamber simulations are compared withdifferent copper structure simulations to quantify thermal performance enhancement.
2. Interposer thermal modeling and its length scaleAs shown in Figure 1, interposers consist of microbumps, interposer substrate, TPVs, andbumps. Top and bottom surfaces of the interposer substrate are laminated with dielectric
Printed Circuit Board
Chip 1 Chip 2
Interposer
�bumps
Through Package Vias (TPVs)(10 ~ 150�m) Bumps (10~400�m)
Redistribution layer (RDL)
Copper Traces (3 ~ 103�m)
Copper Traces
Note: Numbers in brackets are minimum and maximum sizes of the features in each
component
Figure 1.Schematics of 2.5D
glass interposerstructure
1159
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
layers, which work as buildup layers for wiring, referred to as RDLs. Complicated coppertraces are buried in the layer and connected to enable communication between differentchips mounted on the interposer. There are TPVs in the interposer substrate, which passcompletely through the substrate for vertical electrical connection between chips andpackage substrate. The TPVs are either partially or fully filled with copper.
Length scales used for modeling 2.5D interposer range from several micrometers totens of meter. Figure 1 and Table I compare the different length scales used in interposermodeling. When including all the fine details of interposers in the model results in need ofextensive computational time and resources to solve the simulation. As such, a compact orreduced order modeling methodology is needed. Over-simplification of the geometricfeatures, however, can produce large errors in the temperature profile. For this reason,compact modeling of interposer needs a balance between accuracy and computationalefficiency. In this paper, different compact modeling schemes are used for differentinterposer components to develop computationally efficient and accurate thermal model.
3. Compact model for microbump/TPV/Bump modelingMicrobump/TPV/bump arrays are modeled by simplifying the geometry into anequivalent block with effective thermal conductivity. Figure 2 summarizes the compactmodeling procedure used for TPV array. To begin with, a unit cell of TPV is chosen,and thermal boundary conditions to calculate effective thermal conductivity areapplied to this cell. Out-of-plane (z direction) effective thermal conductivity is calculatedby imposing uniform heat flux condition at top and negative heat flux at bottom
Size (width(mm)× length(mm)× height (mm))
Chip 10× 10× 0.5Interposer 25× 25× 0.2Printed circuit board (PCB) 50× 50× 1.2Note: Geometric dimensions of chip, interposer and printed circuit board (PCB)
Table I.Sizes used ininterposer modeling
TPV array in interposerTPV unit cell for
effective thermal conductivity
Heat flux in
Heat flux out
Equivalent block with effective thermal conductivity
Interposer with dielectric layer
TPV array
Simplified TPV array with effective
thermal conductivity
Interposer with dielectric layer
Note: Upper part of TPV array is connected to microbumps, and lower part is connected to
bumps
Figure 2.Compact modelingscheme used forTPV modeling
1160
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
surface, while surrounding surfaces are set as adiabatic. The average temperatures ofthe top and bottom surfaces are obtained and consequently the equivalent thermalconductivity is calculated using:
kef f ¼ Q00DxDT
(1)
where Q00 is the heat flux, Δx the thickness of the sample, and ΔT the temperaturedifference across it. In-plane (x-y direction) effective thermal conductivity is calculatedby applying the same boundary conditions to two side walls, while the otherboundaries are kept as adiabatic. This approach averages local hot spot temperature,which may underestimate the peak temperature. However, using average dietemperature is still a valid approach to evaluate and compare thermal characteristics ofelectronics packaging while a uniform heat generation is assumed. The approach wasvalidated under uniform heat generation boundary condition by comparing thesimplified model with detailed model, which showed only ~1 percent differencebetween maximum temperatures predicted by the two models (Cho et al., 2013). Moredetailed validation of the approach under various geometric conditions is provided inthe literature (Ma et al., 2014). Material properties and the calculated effective thermalconductivities of interconnects and TPV s are shown in Table II.
4. Compact model for copper traces in RDLsIn a RDL, copper traces are patterned in a sophisticated way to enable I/O layout andthe fan out from the chips to a looser pitch footprint. Such redistribution requires thinfilm polymers such as dielectric layers, and metallization to enlarge the pitch of the chipI/Os to match that of another array configuration with larger pitch. Figure 3 shows thetop view of the example four metal layers with different patterns and cross-sectionalview of TPV and copper traces patterned in a package substrate with single IC chip.
Figure 4 explains how in-plane effective thermal conductivity is calculated for metallayer with copper traces when the layers are assumed to be orthotropic. First, the CADdrawing of a metal layer is converted to a binary image, which contains the size andlocation information of the copper traces and the polymer. A black pixel in the figurerepresents copper, and the white pixel represents polymer. Then, the layout is divided intosmall tiles and each tile’s total thermal resistance is obtained by using thermal resistancenetwork analysis. Out-of-plane thermal conductivity of metal layer is calculated with theassumption that black and white pixels are thermally parallel. Pixel resolution of the
kx‒y (W/m K) kz(W/m K)
Silicon 130 130Glass 1 1Polymer 0.8 0.8Underfill material 0.7 0.7Copper 400 400PCB (FR4) 0.3 0.3Microbumps 2.8 4.4Bumps 1.8 4TPV unit cell (glass) 1.2 33.7TPV unit cell (silicon) 119.8 87.3
Table II.Material properties
and calculatedeffective thermalconductivities of
interconnectsand TPVs
1161
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
image is considered to be acceptable when the difference between calculated effectivethermal conductivity from current image and the highest resolution image that CADprogram could export is less than 5 percent. Generally, 25 times reduction of pixels frommaximum resolution results in ~4 percent difference between the results.
Equations (2) and (3) are used to calculate in-plane (row: x-direction, column:y-direction) and out-of-plane (z direction) effective thermal conductivity, respectively:
Rrow=column ¼Number of white pixelsð Þ � Pixel sizeð Þ
kpolymerArow=column
þ Number of black pixelsð Þ � Pixel sizeð ÞkcopperArow=column
Rtotal ¼1P
1=Rrow=column� �
kin�plane ¼Ltotal
Atotal
X1=Rrow=column� �
(2)
White pixel (polymer)
Black pixel (copper)
Copper trace layout (Top view) Zoomed view ofsingle tile
Total resistance
Thermal resistance network for single tile
zx
y
Figure 4.Determination oftotal thermalresistance to obtainin-plane effectivethermal conductivityfor compactmodeling ofmetal layer
Glass
Polymer
Polymer
Polymer
Polymer
Passivation layer10
Thickness (�m)
7 (8 ~ 5)
17.57 (8 ~ 5) 10
17.5
100
17.5
1017.57 (8 ~ 5)
7 (8 ~ 5)10 Passivation layer
Metal layer 1 Cu
TPVCu
Cross-sectional view
Metal layer 2 Cu
Metal layer 3 Cu
Metal layer 4 Cu
Top view
Metal layer 1 Metal layer 2 Metal layer 3 Metal layer 4IC chip
Under fill resin
Build-up Resin(Insulator)
Solder bump (example)
Micro ViaCore resin
Note: Top view of example metal layers’ layout and cross-sectional view of TPV and metal
layers (right)
Source: Left image from www.shinko.co.jp/english/product/buildup/dll.html
Figure 3.Package substratewith single chip
1162
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
Rtotal ¼1
1=Rcopperþ1=Rpolymerkout�of�plane ¼
Acopper
Atotal� kcopperþ
Apolymer
Atotal� kpolymer (3)
The tile with complicated copper traces is then converted into a simple block withcalculated thermal conductivities along x, y, and z direction. More details and thevalidation of this approach are presented in Blackmore (2009).
5. The effect of copper structures on glass and silicon interposersUsing the compact modeling schemes introduced above, the thermal models for 2.5Dglass and silicon interposers are developed. Assuming that chips on the interposers aregenerating the same amount of heat (3W) uniformly, and one ground plane is embeddedin PCB, thermal performance of glass, and silicon interposers with different copperstructures are compared. During the analysis, PCB bottom surface temperature is fixed at300 K and other surfaces are assumed to be adiabatic. Four microbumps under the chipcenter area are assumed to be connected to a single TPV, and four TPVs are connected tosingle bump. Microbumps in the chip periphery area have finer pitch than those in centerarea to provide dense interconnection between two chips. Solder bumps are uniformlydistributed at the bottom of interposer. Four metal layers are incorporated in the model,and identical metal layouts are used for both of the interposers. The schematic of theinterposer structure and metal layer layouts are shown in Figure 5, and detailedgeometric conditions for simulation are provided in Table III.
Interposer
Microbumps
TPVs
Bumps
RDL
Copper Traces
Copper Traces
Chip 2Chip 1
PCB viasGround Plane
Chip center
····
····
····
····
Chip periphery
Interposer
····
····
····
Microbumps–diameter: 46�m, pitch: 80�m
Microbumps–diameter: 46�m, pitch: 150�m
Bumps–diameter: 440�m, pitch: 600�m
TPVs–diameter: 100�m, pitch: 300�m
Figure 5.The schematic of
2.5D interposer (top)microbump/bump/
TPV layout (middle)and layouts of fourmetal layer (bottom)used for simulation
1163
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
Figure 6 compares junction to board thermal resistance of glass and silicon interposerswith different copper structure conditions at given boundary condition. The junction toboard thermal resistance is determined by:
Yjb ¼Tj�Tb
Q(4)
where Tj is junction temperature, Tb is board bottom temperature, and Q is total heatgeneration rate from the chips. Table IV presents four different copper structureconditions applied when modeling TPV, RDL, and PCB. The thermal conductivities ofRDL presented in cases 3 and 4 are the averaged values of thermal conductivities fromfour different metal layers in RDL. Incorporation of copper TPVs and copper ground layerimproves thermal performance of glass significantly, making its out-of-plane effectivethermal conductivity ~34 times higher than that of bare glass. Due to high thermalconductivity of silicon, the effect of TPVs and copper ground layer is not as significant insilicon as compared to the glass interposer case. Increasing in-plane effective thermal
Length (mm) Width (mm) Thickness ( μm) CountChips 10 10 500 2Interposer 25 25 200 1PCB 50 50 1,000 1Ground Plane 50 50 18 1
Dimensions CountMicrobumps ϕ: 46 μm, H: 33 μm Center: 3,600 Periphery: 1,900TPVs ϕ: 100 μm, H: 145 μm 900Bumps ϕ: 440 μm, H: 270 μm 2,500PCB vias ϕ: 400 μm, H: 700 μm 180
Table III.Geometricdimensions ofinterposercomponents
8
7
6
5
4
421Case Number
TPV+
Copper Layer
Copper Tracesin RDL & PCB
Number ofPCB vias
Silicon Interposer
Glass Interposer
Junc
tion
to B
oard
The
rmal
Res
ista
nce
(K/W
)
3
3
Note: The factors that affect the thermal resistance are indicated
on the plot
Figure 6.Junction to boardthermal resistance offour different cases
1164
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
conductivities of RDL and PCB, or out-of-plane effective thermal conductivity of PCB byincreasing number of PCB vias does not increase the performance of both interposersmuch. It is observed that the copper structure implementation has a significant effect onthe thermal performance of glass interposer than that of silicon interposer.
6. Glass interposer with vapor chamber integrated PCBIn this part of the paper, the application of a vapor chamber to interposer structure isdiscussed and its effect on thermal performance is simulated. Figure 7 introduces theconcept of vapor chamber integrated PCB for the interposer. A cavity formed in PCB isplated with copper and is covered with a copper lid, which is directly connected toPCBs. Heat generated from the chips flows through copper TPVs, and heats upthe chamber’s evaporator. The working fluid inside the chamber vaporizes at theevaporator, and limits temperature rise of the device. At the condenser, the vaporcondenses back to liquid, which is attached to a large copper layer heat sink. A wickstructure is used to provide the capillary action needed to drive the liquid againstgravity. Thin layer of wick structure over evaporator area will potentially enhanceliquid supply to the evaporator.
To develop a simplified vapor chamber integrated interposer model, the entire vaporchamber is modeled as a block with an effective thermal conductivity by using the thermalresistance of a vapor chamber with a similar geometry and boundary conditions obtainedfrom literature (Ranjan et al., 2012). An analytical expression for spreading resistance of a
TPV RDL PCB
Case 1 No TPVs No copperkxy, kz¼ 1 W/m K
kxy¼~14W/m KNo. of PCB vias: 180
Case 2 Table III No copperkxy, kz¼ 1 W/m K
kxy¼~14W/m KNo. of PCB vias: 180
Case 3 Table III kx¼ 40W/m Kky¼ 50 W/m Kkz¼ 12 W/m K
kxy¼ 30W/m KNo. of PCB vias: 180
Case 4 Table III kx¼ 40W/m Kky¼ 50W/m Kkz¼ 12W/m K
kxy¼ 30W/m KNo. of PCB vias: 360
Table IV.TPV, RDL, and PCBconditions for four
different cases
Interposer
Microbumps
TPVs
Bumps
RDL
Copper Traces
Copper Traces
Chip 2Chip 1
PCB vias
Vapor Core
Evaporator
Condenser
Heat Flux
h, TfWick
Figure 7.Interposer withvapor chamberintegrated PCB
1165
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
three dimensional rectangular plate (30 mm× 30 mm× 1mm) with single heat source(5 mm× 5mm) shown in Figure 8 is utilized to extract effective thermal conductivity of avapor chamber. To further simplify the model, the effective thermal conductivity of thevapor chamber structure was considered to be isotropic. The governing equation forthe system shown in Figure 8 and its boundary conditions are:
@2T@x2
þ@2T@y2
þ@2T@z2
¼ 0 (5)
@T@z
����z¼0
¼ �Q=As
kef fwithin heat souce area: As ¼ cdð Þ (6)
@T@z
����z¼0
¼ 0 outside the heat source areað Þ (7)
@T@z
����z¼t1
¼ � hkef f
T x; y; t1ð Þ�Tf� �
(8)
@T@x
����x¼0;a
¼ @T@y
����y¼0;b
¼ 0 (9)
The solution for the above differential equations can be obtained by using separation ofvariables. By integrating the solution, mean source temperature can be found as(Yovanovich et al., 1999):
y ¼ y1DþyS ¼ y1Dþ2X1
m¼1Am
cos lmXcð Þ sin 12 lmc� �
lmc
þ2X1
m¼1An
cos dnY cð Þ sin 12 dnd� �
dnd
þ4X1
m¼1
X1n¼1
Amncos dnY cð Þ sin 1
2 dnd� �
cos lmXcð Þ sin 12 lmc� �
lmcdnd(10)
keff
Q
t1
h, Tf
c
db
a
yc
xcx
y
x
z
Figure 8.Isotropic plate withrectangular heatsource on top andboundary conditionsfor analyticalexpression ofthermal resistance
1166
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
where l ¼ mp=a; d ¼ np=b; b ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffil2þd2
p;
Am ¼2Q sin 2Xcþ cð Þ
2 lm�
� sin 2Xc�cð Þ2 lm
� h iabckef f l
2mf lmð Þ (11)
An ¼2Q sin 2Yc þdð Þ
2 dn�
� sin 2Yc�dð Þ2 dn
� h iabdkef f d
2nf dnð Þ (12)
Amn ¼16Q cos lmXcð Þ sin 1
2 lmc� �
cos dnY cð Þ sin 12 dnd� �
abcdkef fbm;nlmdnf bm;n
� � (13)
f zð Þ ¼ zsinh zt1ð Þþh=kef f cosh zt1ð Þzcosh zt1ð Þþh=kef f sinh zt1ð Þ (14)
where ξ¼ λ, δ, or β, and:
y1D ¼ Qab
t1kef f
þ1h
�(15)
Finally, total thermal resistance of vapor chamber can be expressed as:
RVC ¼ yQ¼ y1DþyS
Q¼ R1DþRS (16)
which can be expressed as a function of keff. By equating Equation (16) with the vaporchamber thermal resistance listed in the literature (Ranjan et al., 2012), equivalentthermal conductivity keff was calculated. To find a solution, the Levenberg-Marquardtmethod, a damped least square minimization technique was used. After calculating theeffective thermal conductivity, it was then used for developing finite element (FE)model of a simplified vapor chamber. After FE simulation, thermal resistance wascalculated and the result was compared with the original thermal resistance value fromthe literature for validation, and two thermal resistance values showed ~2 percentdifference. Upon validation, this effective thermal conductivity was then applied to thesimplified block in PCB and used for performance estimation. Figure 9 summarizes thesteps taken for simulation of interposer with vapor chamber embedded PCB.
Two thermal resistance values in the literature (Ranjan et al., 2012), calculated from1D resistance network analysis and numerical model, were used to calculate effectivethermal conductivity of vapor chamber. The result is shown in Table V. Thermalresistance value obtained from the numerical model was used for effective thermalconductivity calculation, as the result from 1D resistance network analysis does notcapture the vapor core resistance and underestimates the total resistance value.
Figure 10 shows the effect of vapor chamber and copper structures studied in thispaper by including the thermal structures sequentially and comparing the thermalperformance of glass interposer to that of silicon interposer for each of these cases.Different TPV, RDL, and PCB conditions used for cases 1, 2, and 3 can be found inTable IV. The previous result for case 4 in Figure 6 is now replaced by the result
1167
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
Vapor chamber thermal resistance data from literature
llAnalytical expression of
thermal resistance
Calculate keff usingLM algorithm
Develop simplified vapor chamber finiteelement model with calculated keffand calculate thermal resistance
Incorporate simplified vapor chamber model in developed interposer model
Compare thermal resistance for validation
Apply validated keff
Microbumps
TPVs
Bumps
RDL
Copper Traces
Copper Traces
Chip 2Chip 1
PCB viasBlock with effective thermal conductivity
Interposer
Figure 9.Thermalperformancesimulation steps forinterposer withvapor chamberintegrated PCB
Thermal resistance (K/W) Effective thermal conductivity (W/m K)
1D resistance analysis 0.27 1,526Numerical model 0.51 491Note: Calculated effective thermal conductivity based on thermal resistance
Table V.Effective thermalconductivity ofvapor chamber
1 2 3 40
1
2
3
4
5
6
7
8
Junc
tion
to B
oard
The
rmal
Res
ista
nce
(K/W
)
Case Number
Glass InterposerSilicon Interposer
TPV+
Copper Layer
Copper Tracesin RDL & PCB
Vapor Chamber
Note: The factors that affect the change of thermal resistance are
indicated on the plot
Figure 10.Junction to boardthermal resistance offour different cases
1168
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
from vapor chamber integrated interposer and shown in Figure 10. After theimplementation of vapor chamber in both interposers, thermal resistance of glassinterposer for case 4 is almost identical to silicon interposer, while the differencebetween the two interposers is significant for case 1. The vapor chamber in PCB, whichprovides better heat spreading effect than thin copper ground layer in PCB, offerssignificant thermal performance enhancement to glass interposer with thermal pathsmade by copper structures.
7. ConclusionsThis study investigates the thermal performance enhancement of glass and siliconinterposers through FE modeling. A 2.5D interposer model was developed to comparethe effect of copper structures on silicon and glass interposers. Detailed interconnectand TPV geometry were incorporated to develop a more realistic model. Major resultsfrom the parametric design study are as follows:
(1) Different compact modeling schemes were used to address multi-scale modelingchallenges in interposer modeling.
(2) The implementation of copper TPVs in interposer and copper ground layer inPCB, enhanced the thermal performance of interposers. TPVs work as thermal,as well as electrical paths.
(3) Comparing the effect of copper TPVs on glass and silicon interposers, TPVs inglass interposer work better as an effective thermal path than in silicon interposer.Glass isolates heat within the chip due to its low thermal conductivity.Silicon interposer substrate, however, spreads the heat which lowers the chiptemperature efficiently.
(4) Thermal performance improvement affected by copper traces in RDL and PCBwas studied. The thermal performance of glass and silicon interposers isaffected more by the in-plane thermal conductivity of PCB, than that of RDL.A change in in-plane thermal conductivity of RDL on silicon interposer hasnegligible effect on the thermal performance of the interposer. Overall,enhancements in out-of-plane thermal conductivity more significantly affectthermal performance, than that in the in-plane thermal conductivity.
(5) Further improvement in thermal performance can be achieved through theimplementation of vapor chamber in PCB. Glass and silicon interposers showalmost identical performance with vapor chamber, overcoming the low thermalconductivity of glass.
(6) During the simulation, both silicon and glass interposers were assumed to havethe same geometric features and conditions. However, as shown in Figure 11,whenever silicon is used as interposer, an additional packaging layer needs tobe added to address thermal stress caused by coefficient of thermal expansionmismatch, which adds an additional thermal resistance to the system. Whenglass interposer is used, there is no need for this additional layer, which canmake thermal performance of glass interposer comparable or better than siliconinterposer with vapor chamber. This also gives more compactness in packagingto glass interposer, making it more suitable for small form factor electronicsystems, than silicon interposer.
1169
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
References
Blackmore, B. (2009), “Validation and sensitivity analysis of an image processing technique toderive thermal conductivity variation within a printed circuit board”, SemiconductorThermal Measurement and Management Symposium (SEMI-THERM 2009), San Jose, CA,March, pp. 76-86.
Cho, S., Joshi, Y., Sundaram, V., Sato, Y. and Tummala, R. (2013), “Comparison of thermalperformance between glass and silicon interposers”, IEEE 63rd Electronic Components andTechnology Conf. (ECTC), Las Vegas, NV, May, pp. 1480-1487.
Heinig, A., Fischbach, R. and Dittrich, M. (2014), “Thermal analysis and optimization of 2.5D and3D integrated systems with wide I/O memory”, 2014 IEEE Intersociety Conference onThermal and Thermomechanical Phenomena in Electronic Systems (ITherm), Orlando, FL,May, pp. 86-91.
Lau, J. and Yue, T.G. (2009), “Thermal management of 3D IC integration with TSV (ThroughSilicon Via)”, IEEE Electronic Components and Technology Conference (ECTC), San Diego,CA, May, pp. 635-640.
Ma, H., Yu, D. and Wang, J. (2014), “The development of effective model for thermal conductionanalysis for 2.5D packaging using TSV interposer”, Microelectronics Reliability, Vol. 54No. 2, pp. 425-434.
Oprins, H. and Beyne, E. (2014), “Generic thermal modeling study of the impact of 3D-interposermaterial and thickness options”, 2014 IEEE Intersociety Conference on Thermal andThermomechanical Phenomena in Electronic Systems (ITherm), Orlando, FL, May,pp. 72-78.
Oprins, H., Srinivasan, A., Cupak, M., Cherman, V., Torregiani, C., Stucchi, M., Van der Plas, G.,Marchal, P., Vandevelde, B. and Cheng, E. (2011), “Fine grain thermal modelingand experimental validation of 3D-ICs”, Microelectronics Journal, Vol. 42 No. 4,pp. 572-578.
(a)
(b)
Package Substrate
Chip 1 Chip 2
Interposer
Microbumps
TPVs
Solder Balls
RDL
Copper Traces
Copper Traces
Printed Circuit Board
Bumps
Printed Circuit Board
Chip 1 Chip 2
Interposer
Microbumps
TPVs
Bumps
RDL
Copper Traces
Copper Traces
Figure 11.(a) Silicon interposergeometry withadditional packaginglayer, and (b) Glassinterposer geometry
1170
HFF26,3/4
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)
Ranjan, R., Murthy, J.Y., Garimella, S.V., Altman, D.H. and North, M.T. (2012), “Modeling anddesign optimization of ultrathin vapor chambers for high heat flux applications”, IEEETransactions on Components, Packaging and Manufacturing Technology, Vol. 2 No. 9,pp. 1465-1478.
Sukumaran, V., Chen, Q., Liu, F., Kumbhat, N., Bandyopadhyay, T., Hunter, C., Min, S., Nopper, C.,Sundaram, V. and Tummala, R. (2010), “Through-package-via formation and metallizationof glass interposers”, IEEE 60th Electronic Components and Technology Conf. (ECTC),Las Vegas, NV, June, pp. 557-563.
Sundaram, V., Sato, Y., Seki, T., Takagi, Y., Smet, V., Kobayashi, M. and Tummala, R. (2014),“First demonstration of a surface mountable, ultra-thin glass BGA package for smartmobile logic devices”, IEEE Electronic Components and Technology Conference (ECTC),Orlando, FL, May, pp. 365-370.
Yovanovich, M.M., Muzychka, Y.S. and Culham, J.R. (1999), “Spreading resistance of isofluxrectangles and strips on compound flux channels”, Journal of Thermo-Physics and HeatTransfer, Vol. 13 No. 4, pp. 495-500.
Zhang, H.Y. and Zhang, X.W. (2015), “Study of thermally enhanced 2.5D packages with multi-chipsmolded on silicon interposer”, Journal of Electronic Materials, Vol. 44 No. 7, pp. 2396-2405.
Zhang, H.Y., Zhang, X.W., Lau, B.L., Lim, S., Ding, L. and Yu, M.B. (2014), “Thermal characterizationof both bare die and overmolded 2.5-D packages on through silicon interposers”, IEEETransactions on Components, Packaging and Manufacturing Technology, Vol. 4 No. 5,pp. 807-816.
Corresponding authorYogendra Joshi can be contacted at: [email protected]
For instructions on how to order reprints of this article, please visit our website:www.emeraldgrouppublishing.com/licensing/reprints.htmOr contact us for further details: [email protected]
1171
Multi-scalethermal
modeling
Dow
nloa
ded
by G
eorg
ia I
nstit
ute
of T
echn
olog
y A
t 08:
37 2
9 A
ugus
t 201
6 (P
T)